The Automatic Image Registration Algorithm Based on Bag of Words and Features Point

Abstract

According to actual requirements and situation, a novel automatic image registration algorithm was presented based on bag of words and features point. Firstly, pyramid delaminating was used to preprocess the pre-registration images. Then, image normalized could be obtained. Secondly, the dense SIFT features points could be extracted. The paper could get the feature eigenvectors descriptions of image by k-means clustering algorithm. Thirdly, the feature eigenvectors descriptions of image were train and the reference and pre-registration images were classification through Support Vector Machine (SVM). Finally, image registration could implement by features matching points. The results of experiment with the with office room and building test images shown the proposed algorithm was effective and could obtain the better registration image.